Clean a noisy phone recording: how to make a conversation easier to hear
Identify noise in phone and voice-recorder conversations, compare profile-based and speech-focused treatment, and review the result.
Noise reduction
Reduce hiss, fans and persistent background noise using a learned noise profile or adaptive estimation.
When should you choose this tool?
For a steady wash of noise underneath speech and between phrases, start with noise reduction. A speech-free passage can teach the background character. Adaptive mode follows estimated changes in the noise; an isolated click or impact is better addressed with De-click or spectral attenuation.
How it changes audio
Processing divides short audio windows into frequency components and attenuates them using the learned or estimated noise spectrum. Basic and Detailed are different processing methods: Detailed Learn requires a profile learned for that method.
What do the parameters control?
- Processing; Learn / Adaptive
- Choose Basic or Detailed. Select only background noise before Learn. Adaptive needs no separate sample, but its estimate can mistake speech for noise.
- Reduction
- Noise attenuation in dB: 0–30 dB in Basic and 0–40 dB in Detailed. Higher settings can thin the voice or make it metallic.
- Threshold offset
- Moves the processing threshold relative to learned or estimated noise. It is distinct from the input trace and from an absolute recording level.
- Frequency smoothing; Release
- In Basic, these control variation between neighbouring frequencies and how attenuation releases over time. Too little smoothing can create musical artifacts; long release can smear detail.
- Frequency resolution
- Basic offers 2048- or 4096-point analysis. A larger window resolves frequencies more finely with less temporal precision. Keep the window matched to the learned profile.
- Temporal smoothing; Residual whitening; Noise estimator
- Detailed controls shape temporal behaviour and the residual noise character. Adaptive offers Basic, Trimmed mean and Minimum statistics estimators. Compare them on the same phrase.
Try a short passage
- Listen to a quiet and a loud phrase. Select a speech-free background passage and use Learn, or stay in Adaptive if none is available.
- Select the speech to process. Start Preview with gentle reduction and compare Result with Input and Residual.
- Change one setting at a time. If consonants or word endings disappear, ease off reduction or threshold before applying the step.
What to listen for
Quiet pauses should come with intact soft syllables. Clearly intelligible speech in Residual indicates that useful voice content was removed.
Limits of the method
Noise overlapping speech in time and frequency cannot be guaranteed to disappear without loss. Speech included in the noise profile may later be treated as noise.
Preview covers a short passage. After applying, review the full treated section; you can undo the step from history.
Related problems and workflows
Reduce reverb in a recording: improve roomy speech
Distinguish room decay from repeated echoes, then set up and review De-reverb.